mlsanityData to 14 Oct 2026

Text2VLM: Adapting Text-Only Datasets to Evaluate Alignment Training in Visual Language Models

arXiv 2507.20704 ↗First posted 28 Jul 2025cs.CL cs.AI cs.CRcited by 0

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

also in 2 more
Abstract (arXiv)

The increasing integration of Visual Language Models (VLMs) into AI systems necessitates robust model alignment, especially when handling multimodal content that combines text and images. Existing evaluation datasets heavily lean towards text-only prompts, leaving visual vulnerabilities under evaluated. To address this gap, we propose Text2VLM, a novel multi-stage pipeline that adapts text-only datasets into multimodal formats, specifically designed to evaluate the resilience of VLMs against typographic prompt injection attacks. The Text2VLM pipeline identifies harmful content in the original text and converts it into a typographic image, creating a multimodal prompt for VLMs. Also, our evaluation of open-source VLMs highlights their increased susceptibility to prompt injection when visual inputs are introduced, revealing critical weaknesses in the current models' alignment. This is in addition to a significant performance gap compared to closed-source frontier models. We validate Text2VLM through human evaluations, ensuring the alignment of extracted salient concepts; text summarization and output classification align with human expectations. Text2VLM provides a scalable tool for comprehensive safety assessment, contributing to the development of more robust safety mechanisms for VLMs. By enhancing the evaluation of multimodal vulnerabilities, Text2VLM plays a role in advancing the safe deployment of VLMs in diverse, real-world applications.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 18 of its 24 references that are in the map. First posted on arXiv 28 Jul 2025.

Builds on

18 of 24 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024Learning To See But Forgetting To Follow: Visual Instruction Tuning Makes LLMs More Prone To Jailbreak Attacks7 May
  2. 2024Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models14 Mar
  3. 2024Safety of Multimodal Large Language Models on Images and Texts1 Feb
  4. 2023MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models29 Nov
  5. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov

Other references · 13

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  2. 2023Improved Baselines with Visual Instruction Tuningcited by 4,418
  3. 2023VILA: On Pre-training for Visual Language Modelscited by 692
  4. 2024Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attackscited by 486
  5. 2023OCRBench: On the Hidden Mystery of OCR in Large Multimodal Modelscited by 476
  6. 2022ToxiGen: A Large-Scale Machine-Generated Dataset for Adversarial and Implicit Hate Speech Detectioncited by 422
  7. 2024Great, Now Write an Article About That: The Crescendo Multi-Turn LLM Jailbreak Attackcited by 390
  8. 2023Purple Llama CyberSecEval: A Secure Coding Benchmark for Language Modelscited by 138
  9. 2024Jailbreaking Attack against Multimodal Large Language Modelcited by 135
  10. 2024Improved Few-Shot Jailbreaking Can Circumvent Aligned Language Models and Their Defensescited by 70
  11. 2024Visual-RolePlay: Universal Jailbreak Attack on MultiModal Large Language Models via Role-playing Image Charactercited by 68
  12. 2022Detecting and Understanding Harmful Memes: A Surveycited by 58
1 more
  1. 2024MedSafetyBench: Evaluating and Improving the Medical Safety of Large Language Modelscited by 44

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026One Perturbation, Two Failure Modes: Probing VLM Safety via Embedding-Guided Typographic Perturbations
  2. 2024VLSBench: Unveiling Visual Leakage in Multimodal Safety
  3. 2026ReViCo: Unveiling the Limitations of VLMs in Visual Text Understanding via Error Correction
  4. 2024ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time
  5. 2026Reading Between the Pixels: Linking Text-Image Embedding Alignment to Typographic Attack Success on Vision-Language Models
  6. 2025Trust the Model: Compact VLMs as In-Context Judges for Image-Text Data Quality
  7. 2024The Revolution of Multimodal Large Language Models: A Survey
  8. 2025VISTA: Enhancing Vision-Text Alignment in MLLMs via Cross-Modal Mutual Information Maximization